A multi-view clustering method based on uncertainty and graph neural network

Abstract

The present invention provides a multi-view clustering method based on uncertainty and graph neural network, which relates to the technical field of multi-view clustering. The present invention obtains a multi-view data set, performs preprocessing, and obtains an original feature matrix and an adjacency matrix; extracts and fuses the input original feature matrix and the adjacency matrix to obtain a consistent embedding matrix and a transfer matrix; based on Dempster‑Shafer evidence theory and Dirichlet distribution, a reliable consistent embedding is obtained by reducing the uncertainty of the embedding space in the consistent embedding matrix; the reliable consistent embedding and the transfer matrix are mapped to a final embedding; the final embedding is decoded, reconstructed and clustered to obtain a clustering result. The present invention can fully tap the potential information in the view, reduce the uncertainty in the model, solve the noise problem existing in the multi-view clustering task, and can quickly, efficiently and accurately obtain the clustering results of multi-view data.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-view clustering, and in particular to a multi-view clustering method based on uncertainty and graph neural network. Background Art

[0002] Multi-view data refers to data sets obtained from different perspectives, different feature extraction methods or different data sources. Multi-view clustering tasks cluster data into corresponding clusters by learning from different perspectives and using the complementary information between multiple perspectives. It is a common method for processing multi-view data and is widely used in social network analysis, image processing, bioinformatics and other fields. Multi-view data originates from different representations of the same object or event, and often contains consistency and complementarity in different views. In order to effectively utilize this information, many studies have proposed multi-view clustering methods. However, due to the complexity of multi-view data, traditional clustering methods are difficult to fully mine the useful information of all views. In addition, during the data set collection process, due to the inevitable noise in multi-view data sets, the stability and accuracy of the clustering results are difficult to guarantee. Therefore, a multi-view clustering method based on uncertainty and graph neural network is introduced to improve the robustness and generalization ability of the model by considering the uncertainty in each view.

[0003] With the popularity of graph neural networks, more and more researchers have begun to apply them to multi-view clustering tasks. Through graph neural networks, graph structures can be built in multi-view data to better model the relationship between views and help mine potential information in views. Multi-view clustering methods based on uncertainty and graph neural networks can fully mine potential information in views and reduce uncertainty in the model, solving the noise problem in multi-view clustering tasks.

[0004] In view of this, the applicant filed this application after studying the existing technology. Summary of the invention

[0005] The present invention aims to provide a multi-view clustering method based on uncertainty and graph neural network to solve the shortcomings of existing methods such as model uncertainty and noise in multi-view clustering tasks.

[0006] In order to solve the above technical problems, the present invention is implemented through the following technical solutions:

[0007] A multi-view clustering method based on uncertainty and graph neural network, including:

[0008] S1, obtain the multi-view dataset, and after preprocessing, obtain the original feature matrix and adjacency matrix;

[0009] S2, extracting and fusing the input original feature matrix and the adjacency matrix to obtain a consistent embedding matrix and a transfer matrix;

[0010] S3, based on Dempster-Shafer evidence theory and Dirichlet distribution, a reliable consistent embedding is obtained by reducing the uncertainty of the embedding space in the consistent embedding matrix;

[0011] S4, mapping the reliable consistent embedding and the transfer matrix into a final embedding;

[0012] S5, performing decoding reconstruction and cluster analysis on the final embedding to obtain a clustering result.

[0013] Preferably, the pretreatment comprises:

[0014] The original feature matrix is ​​obtained by performing feature extraction on the multi-view dataset;

[0015] Based on the Euclidean distance, generating node edges for the nearest neighbors of each node of the original feature matrix to obtain an initial adjacency graph;

[0016] Using a pruning strategy on the initial adjacency graph to obtain a key adjacency graph to improve the quality of the adjacency graph;

[0017] The key adjacency graph is normalized to obtain the adjacency matrix.

[0018] Preferably, S2 is specifically:

[0019] Inputting the original feature matrix and the adjacency matrix into a graph convolution encoder for feature extraction to obtain an initial embedding; wherein the graph convolution encoder includes two graph convolution layers and a ReLU activation function;

[0020] The initial embedding and the adjacency matrix are fused respectively to obtain a consistent embedding matrix and a transfer matrix.

[0021] Preferably, the step S3 is specifically:

[0022] Step 1: Calculate the cosine similarity between the data points of the initial embedding and the consistent embedding matrix. The formula is:

[0023] ;

[0024] in, The i-th data point representing the initial embedding of the v-th view; represents the jth data point of the consistent embedding matrix; T represents transpose;

[0025] Step 2: According to the cosine similarity, the evidence pair of the corresponding data point is obtained, and the formula is:

[0026] ;

[0027] in, Represents data points and The evidence is right; Represents the combined function of Tanh and ReLu;

[0028] Step 3: Calculate the belief quality and uncertainty between the initial embedding and the consistent embedding matrix based on the evidence pair. The formula is:

[0029] ;

[0030] ;

[0031] ;

[0032] in, represents the quality of the belief; express the said uncertainty; Represents the Dirichlet distribution parameters corresponding to data point i and data point j; represents the Dirichlet intensity; N represents the number of samples;

[0033] Step 4: Use the probability density function of Dirichlet distribution to model the uncertainty, the formula is:

[0034] ;

[0035] in, Represents a random variable The obedience parameters are The Dirichlet distribution of represents the overall reliability probability of data point i; represents the reliability probability of the similarity between data point i and data point j, ; represents all Dirichlet distribution parameters associated with data point i; represents the N-dimensional beta function; represents the N-dimensional unit simplex;

[0036] Step 5: obtaining a similarity matrix according to the cosine similarity; obtaining an uncertainty vector according to the uncertainty; constructing an indicator matrix according to the similarity matrix and the uncertainty vector to screen out reliable embedding information and obtain a reliable consistent embedding matrix, i.e., the reliable consistent embedding;

[0037] According to the indicator matrix, through the loss function and Back propagation, continuously optimizing and updating model parameters, the formula is:

[0038] ；

[0039] ;

[0040] in, represents the i-th vector of the indicator matrix; represents the KL divergence measure; represents the parameters of the Dirichlet distribution after removing unreliable evidence from the predicted Dirichlet distribution.

[0041] Preferably, S4 specifically comprises: taking the reliable consistent embedding and the transfer matrix as input of a graph convolutional encoder, passing through two graph convolutional layers and a ReLU activation function, to obtain the final embedding.

[0042] Preferably, S5 is specifically:

[0043] The final embedding is obtained by reconstructing the initial input through the decoder to obtain the reconstructed embedding, and the mean square error loss function is used to optimize the difference between the reconstructed embedding and the reliable consistent embedding. The formula of the mean square error loss function is:

[0044] ;

[0045] in, represents the reliable consistent embedding matrix; represents the reconstructed embedding; F represents the Frobenius norm of the matrix; N represents the number of samples; represents the dimension of fused embedding;

[0046] According to the final embedding, a clustering result is obtained by a clustering function.

[0047] Preferably, the method further comprises: inputting the preprocessed and category-labeled multi-view training set into the multi-view clustering model for training and parameter fine-tuning to obtain a trained multi-view clustering model;

[0048] During the training process, the clustering loss function is used to train the clustering target, the clustering result of the multi-view clustering model is optimized, the initial probability distribution Q is calculated using the t-student distribution, and then the target probability distribution P is calculated based on the initial probability distribution Q. The formula of the clustering loss function is:

[0049] ;

[0050] in, represents the clustering loss function, which represents the difference between the initial probability distribution Q and the target probability distribution P; represents the KL divergence measure; and Respectively indicate that data point i belongs to cluster The target probability P and the initial probability Q.

[0051] The present invention also provides a multi-view clustering device based on uncertainty and graph neural network, comprising:

[0052] The preprocessing unit is used to obtain the multi-view data set and obtain the original feature matrix and adjacency matrix after preprocessing;

[0053] A final embedding unit is used to extract and fuse the input original feature matrix and the adjacency matrix to obtain a consistent embedding matrix and a transfer matrix;

[0054] An uncertainty unit, used for obtaining a reliable consistent embedding by reducing the uncertainty of the embedding space in the consistent embedding matrix based on Dempster-Shafer evidence theory and Dirichlet distribution;

[0055] A final embedding unit, configured to map the reliable consistent embedding and the transfer matrix into a final embedding;

[0056] The output unit is used to perform decoding reconstruction and cluster analysis on the final embedding to obtain a clustering result.

[0057] The present invention also provides a multi-view clustering device based on uncertainty and graph neural network, including a processor and a memory, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a multi-view clustering method based on uncertainty and graph neural network as described above.

[0058] The present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, a multi-view clustering method based on uncertainty and graph neural network is implemented as described above.

[0059] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0060] The method of the present invention extracts features from multi-view data based on graph neural networks, uses evidence theory and Dirichlet distribution to reduce the uncertainty in the embedding space of the data, and uses the obtained reliable consistent embedding matrix to perform subsequent clustering tasks, fully mining the potential information in the views, reducing the uncertainty in the model, and solving the noise problem existing in the multi-view clustering task.

[0061] In addition, the present invention continuously optimizes and updates the model parameters through the loss function and the kl back propagation function, thereby improving the stability and robustness of the multi-view clustering model, making the model have better performance and being able to quickly, efficiently and accurately obtain the clustering results of multi-view data. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 A schematic diagram of a multi-view clustering method based on uncertainty and graph neural network provided in Example 1 of the present invention.

[0064] Figure 2 A flowchart of a multi-view clustering method based on uncertainty and graph neural network provided in Example 1 of the present invention.

[0065] Figure 3 This is a flow chart of the uncertainty module provided in the first embodiment of the present invention.

[0066] Figure 4 A schematic diagram of a multi-view clustering device based on uncertainty and graph neural network provided in Example 2 of the present invention.

[0067] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0069] Embodiment 1

[0070] Embodiment 1 of the present invention provides a multi-view clustering method based on uncertainty and graph neural network, which can be implemented by a multi-view clustering device based on uncertainty and graph neural network (hereinafter referred to as the multi-view clustering device), and in particular, executed by one or more processors in the multi-view clustering device.

[0071] In this embodiment, the multi-view clustering device may be an electronic device equipped with a processor, the processor having a computer program of the multi-view clustering method based on uncertainty and graph neural network and the computer program can be executed, such as a computer, a smart phone, a smart tablet, a workstation, etc., which is not limited here.

[0072] In this embodiment, the multi-view dataset is a data set consisting of multiple sets of different features or description methods, such as social network analysis data, image processing data, and biomedical informatics data, and each set of feature descriptions is called a view. The multi-view dataset contains data from multiple views, and each view provides an observation or understanding of different aspects of the data object. These views can be different types of data, such as numerical data, text data, image data, etc., and may come from different data sources or be obtained through different feature extraction methods. For example:

[0073] Image feature extraction data: Multiple features can be extracted from a picture, such as LBP (local binary pattern), SIFT (scale invariant feature transform), HOG (histogram of oriented gradients), GIST (global feature descriptor), etc. These features can be regarded as different views of the picture.

[0074] Multilingual text data: An article can be described in multiple languages, such as English, Chinese, Japanese, etc. The text content in each language can be regarded as a view of the article.

[0075] Medical imaging data: In the field of medical imaging, images collected by devices such as MRI (magnetic resonance imaging), PET (positron emission tomography), and CSF (cerebrospinal fluid) provide functional information, structural anatomical information, and genetic information of patients. These image data constitute a multi-view dataset.

[0076] Social network data: A user's social network data may include multiple views, such as text view (text content of posts and comments posted by the user), social relationship view (friendship or interaction frequency between users), and behavior view (user's browsing history, likes or sharing behavior, etc.).

[0077] This embodiment aims to provide a multi-view clustering method based on uncertainty and graph neural network to solve the shortcomings of the existing methods, such as the uncertainty of the multi-view clustering model and the existence of noise in the multi-view clustering task.

[0078] like Figure 1 As shown, a multi-view clustering method based on uncertainty and graph neural network includes steps S1 to S4.

[0079] S1, obtain the multi-view dataset, and after preprocessing, obtain the original feature matrix and adjacency matrix.

[0080] Specifically, the multi-view dataset is processed by a graph construction method to obtain the original feature matrix of multiple samples and the adjacency relationship under a specific view, so as to construct the adjacency matrix of the specific view, which specifically includes steps A1 to A3.

[0081] Step A1: extracting features from the multi-view dataset to obtain the original feature matrix, and finding the nearest neighbor of each node in the original feature matrix by Euclidean distance to generate the edge of the node, that is, the initial adjacency graph;

[0082] Step A2: Use pruning strategy to improve the quality of the graph constructed in step A1 and obtain the key adjacency graph;

[0083] Step A3: Normalize the key adjacency graph to obtain the final adjacency matrix.

[0084] In this embodiment, when constructing an adjacency graph, especially when processing large data sets, a pruning strategy can be used to reduce unnecessary computational or storage overhead while maintaining key structural properties of the graph. Pruning usually involves removing or ignoring edges and nodes that have little or no impact on the final result.

[0085] The following pruning strategies can be used when constructing an adjacency graph, such as distance-based threshold pruning, importance-based pruning, local connectivity pruning, and algorithm-based pruning.

[0086] The preprocessed multi-view dataset includes multiple original feature matrices from different data sources or different feature extraction methods, multiple adjacency matrices processed by steps A1 to A3, and the category label corresponding to each data point, which is convenient for model training and testing.

[0087] S2, after extracting and fusing the input original feature matrix and the adjacency matrix, a consistent embedding matrix and a transfer matrix are obtained.

[0088] In this embodiment, the “fusion” refers to multiplying the parameters obtained by adaptive learning with the embedding matrix or transfer matrix of each view and then summing them up to obtain the corresponding target matrix.

[0089] like Figure 2 As shown, in this embodiment, a multi-view clustering model based on uncertainty and graph convolutional neural network is constructed for training, and the multi-view clustering model includes a graph convolutional encoder, an uncertainty module and a decoder, and the graph convolutional encoder includes two graph convolutional layers and a ReLU activation function.

[0090] Step B1: Use the view-specific graph convolutional encoder to extract features from the preprocessed dataset to obtain the initial embedding, specifically:

[0091] The original feature matrix obtained after preprocessing and the adjacency matrix As the input of the graph convolutional encoder; the input and In the graph convolutional encoder, two graph convolutional layers and a ReLU activation function are passed to finally extract the initial view-specific embedding .

[0092] in, Represents the vth view.

[0093] Step B2: Initial Embed The consistent embedding matrix H is obtained by fusion. Meanwhile, the adjacency matrices of multiple views are fused to obtain the transfer matrix U.

[0094] in, represents the i-th initial embedding, .

[0095] The “fusion” mentioned here refers to multiplying the parameters obtained by adaptive learning with the embedding matrix of each view and then summing them up to finally obtain a consistent embedding matrix.

[0096] S3, based on Dempster-Shafer evidence theory and Dirichlet distribution, a reliable consistent embedding is obtained by reducing the uncertainty of the embedding space in the consistent embedding matrix.

[0097] Step B3: Figure 3 As shown, the uncertainty of the embedding space is reduced through the uncertainty module of the model, and a more reliable consistent embedding matrix is ​​obtained. Specifically, it includes steps B31 to B35.

[0098] Step B31: Calculate the cosine similarity of data points between the initial embedding and the consistent embedding matrix ;

[0099] in, The i-th data point representing the initial embedding of the v-th view; represents the jth data point of the consistent embedding matrix; T represents transpose;

[0100] Step B32: In order to estimate the uncertainty in the embedding space, the uncertainty in the embedding space is modeled according to the Dempster-Shafer evidence theory (ie, the belief function theory or DS theory).

[0101] According to cosine similarity Obtain evidence for:

[0102] ;

[0103] in, Represents data points and The evidence is that Represents the combined function of Tanh and ReLu.

[0104] Step B33: Use evidence to calculate and the quality of beliefs between them and their uncertainty;

[0105] ;

[0106] ;

[0107] ;

[0108] in, represents the quality of the belief; Indicates uncertainty; Represents the Dirichlet distribution parameters corresponding to data point i and data point j; represents the Dirichlet intensity; N represents the number of samples.

[0109] Next, the Dirichlet distribution parameterized by evidence is used to model uncertainty. The probability density function of the Dirichlet distribution is as follows:

[0110] ;

[0111] in, Represents a random variable The obedience parameters are The Dirichlet distribution of Reliability probability indicating similarity; Represents the reliability probability of the similarity between data point i and data point j; represents the Dirichlet distribution parameter of data point i; represents the N-dimensional beta function; represents the N-dimensional unit simplex.

[0112] In this embodiment, the N-dimensional unit simplex represents a convex hull consisting of a set of N+1 affine-independent points in N-dimensional space, and the coefficients of the convex combination of these points satisfy the conditions of the unit simplex.

[0113] Step B34: Obtain the similarity matrix using the calculation method of similarity cosine degree in step B31 ; Use the uncertainty calculation method of step B33 to obtain the uncertainty vector ; Using the similarity matrix and the uncertainty vector Constructing an indicator matrix , in order to filter out reliable embedding pair information.

[0114] Step B35: Calculate the loss function and conduct Backpropagation training model, backpropagation is used to calculate the gradient of the loss function with respect to the model parameters, and update the parameters based on these gradients. The loss function and backpropagation function are expressed as:

[0115] ;

[0116] ;

[0117] in, represents the i-th vector of the indicator matrix; represents KL divergence; represents the Dirichlet parameter after removing unreliable evidence from the predicted Dirichlet distribution.

[0118] S4, mapping the reliable consistent embedding and the transfer matrix into a final embedding.

[0119] In this step, the obtained reliable and consistent embedding matrix and transfer matrix are mapped to the final embedding using a graph convolutional encoder.

[0120] Specifically, the reliable consistent embedding matrix obtained in step B3 and the transfer matrix obtained in step B22 are used as the input of the graph convolution encoder. The data will pass through two graph convolution layers and a ReLU activation function to obtain the final embedding.

[0121] S5, performing decoding reconstruction and cluster analysis on the final embedding to obtain a clustering result.

[0122] In this step, the final embedding is decoded and reconstructed by the decoder to obtain the reconstructed embedding, and the mean square error loss function is used to optimize the difference between the two. The expression of the mean square error loss function is:

[0123] ;

[0124] in, represents the reliable consistent embedding matrix; represents said reconstructed embedding; represents the Frobenius norm of the matrix; N represents the number of samples; Represents the dimension of the fused embedding.

[0125] Then, the structural loss of the final embedding is calculated, and the final embedding is passed through a clustering function to obtain a final clustering result.

[0126] In this embodiment, the clustering function may adopt an existing clustering function, such as k-means clustering, K-center point clustering, and the like.

[0127] This embodiment also includes: inputting the preprocessed and category-labeled multi-view training set into the constructed multi-view clustering model for model training, adjusting the parameters of the model, and saving the model with adjusted parameters to obtain a trained multi-view clustering model.

[0128] During the training process, a clustering loss function is used to self-train the clustering target and optimize the clustering result of the multi-view clustering model. In this embodiment, the initial probability distribution Q is calculated using t-student distribution, and then the target probability distribution P is calculated based on the initial probability distribution.

[0129] Clustering loss function is usually used to measure the quality of clustering results. In clustering methods based on probability distribution, clustering loss function can be defined as the difference between the initial probability distribution Q and the target probability distribution P. A commonly used measurement method is KL (Kullback-Leibler Divergence). Of course, according to actual conditions, other measurement methods can also be used, such as Euclidean distance, Manhattan distance, Chebyshev distance, etc., which are not limited here.

[0130] In this embodiment, the calculation formula of the clustering loss function combined with KL divergence is as follows:

[0131] ;

[0132] in, represents the clustering loss function; represents KL divergence; and Respectively indicate that data point i belongs to cluster The target probability P and the initial probability Q.

[0133] In this embodiment, when testing the model, a public multi-view dataset is used for testing. For all experiments, the same hardware environment is guaranteed. The training dataset uses the public dataset BBCSport, which contains 544 samples and document data of five topics, with two views and dimensions of 3183 and 3203 respectively.

[0134] This embodiment uses a graph construction method during the training process, and generates the edges of the nodes by calculating the Euclidean distance to find the nearest neighbor of each node.

[0135] In this embodiment, three common evaluation indicators are used to evaluate the model test results: Clustering accuracy (ACC), Normalized Mutual Information (NMI) and Adjusted Rand Index (ARI). When the model evaluation result reaches the set threshold, the trained multi-view clustering model is obtained.

[0136] The acquired new multi-view dataset is input into the trained multi-view clustering model after preprocessing to obtain the multi-view clustering result.

[0137] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0138] The method of the present invention extracts features from multi-view data based on graph neural networks, uses evidence theory and Dirichlet distribution to reduce the uncertainty in the embedding space of the data, and uses the obtained reliable consistent embedding matrix for subsequent clustering tasks, which fully mines the potential information in the views, reduces the uncertainty in the model, and solves the noise problem existing in the multi-view clustering task.

[0139] In addition, the present invention continuously optimizes and updates the model parameters through the loss function and the kl back propagation function, thereby improving the stability and robustness of the multi-view clustering model, making the model have better performance and being able to quickly, efficiently and accurately obtain the clustering results of multi-view data.

[0140] Embodiment 2

[0141] like Figure 4 As shown, the second embodiment of the present invention further provides a multi-view clustering device based on uncertainty and graph neural network, including:

[0142] A preprocessing unit, used to obtain a multi-view data set, and perform preprocessing to obtain a preprocessed multi-view data set; wherein the preprocessed multi-view data set includes an original feature matrix and an adjacency matrix;

[0143] A final embedding unit is used to extract and fuse the input original feature matrix and the adjacency matrix to obtain a consistent embedding matrix and a transfer matrix;

[0144] An uncertainty unit, used for obtaining a reliable consistent embedding by reducing the uncertainty of the embedding space in the consistent embedding matrix based on Dempster-Shafer evidence theory and Dirichlet distribution;

[0145] A final embedding unit, configured to map the reliable consistent embedding and the transfer matrix into a final embedding;

[0146] The output unit is used to perform decoding reconstruction and cluster analysis on the final embedding to obtain a clustering result.

[0147] Embodiment 3

[0148] The third embodiment of the present invention also provides a multi-view clustering device based on uncertainty and graph neural network, and applies the method of the present invention to a multi-view clustering system based on uncertainty and graph neural network, which includes a memory and a processor, and the memory stores a computer program, and the computer program can be executed by the processor to implement the multi-view clustering method based on uncertainty and graph neural network as mentioned above.

[0149] Embodiment 4

[0150] The fourth embodiment of the present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, the multi-view clustering method based on uncertainty and graph neural network as described above is implemented.

[0151] In several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method can also be implemented in other ways. The apparatus and method embodiments described above are merely schematic. For example, the flowcharts in the accompanying drawings show the possible architecture, functions and operations of the apparatus, method and computer program product according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0152] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0153] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0154] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0155] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0156] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0157] The "first\second" mentioned in the embodiments is only to distinguish similar objects, and does not represent a specific order for the objects. It is understandable that the "first\second" can be interchanged with the specific order or sequence where permitted. It should be understood that the objects distinguished by "first\second" can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than those illustrated or described herein.

[0158] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-view clustering method based on uncertainty and graph neural network, characterized in that: include: S1, obtain the multi-view dataset, and after preprocessing, obtain the original feature matrix and adjacency matrix; S2, extracting features from the original feature matrix and the adjacency matrix of the input graph convolutional encoder to obtain an initial embedding; The initial embedding and the adjacency matrix are respectively fused to obtain a consistent embedding matrix and a transfer matrix; S3, based on Dempster-Shafer evidence theory and Dirichlet distribution, by reducing the uncertainty of the embedding space in the consistent embedding matrix, a reliable consistent embedding is obtained, specifically including: Step 1: Calculate the cosine similarity between the data points of the initial embedding and the consistent embedding matrix ; Step 2: According to the cosine similarity, the evidence pair of the corresponding data point is obtained, and the formula is: ; in, Indicates the i-th data point of the initial embedding of the view; represents the j-th data point of the consistent embedding matrix; Represents data points and The evidence is right; Represents the combined function of Tanh and ReLu; Step 3, calculating the belief quality and uncertainty between the initial embedding and the consistent embedding matrix according to the evidence pair; Step 4, using the probability density function of Dirichlet distribution to model the uncertainty; Step 5: obtaining a similarity matrix according to the cosine similarity; obtaining an uncertainty vector according to the uncertainty; constructing an indicator matrix according to the similarity matrix and the uncertainty vector to screen out reliable embedding information and obtain a reliable consistent embedding matrix, i.e., the reliable consistent embedding; S4, mapping the reliable consistent embedding and the transfer matrix into a final embedding based on a graph convolutional encoder; S5, performing decoding reconstruction and cluster analysis on the final embedding to obtain a clustering result.

2. According to claim 1, a multi-view clustering method based on uncertainty and graph neural network is characterized in that , the preprocessing includes: The original feature matrix is ​​obtained by performing feature extraction on the multi-view dataset; Based on the Euclidean distance, generating node edges for the nearest neighbors of each node of the original feature matrix to obtain an initial adjacency graph; Using a pruning strategy on the initial adjacency graph to obtain a key adjacency graph to improve the quality of the adjacency graph; The key adjacency graph is normalized to obtain the adjacency matrix.

3. The multi-view clustering method based on uncertainty and graph neural network according to claim 1 is characterized in that ,The graph convolutional encoder includes two graph convolutional layers and a ReLU activation function.

4. The multi-view clustering method based on uncertainty and graph neural network according to claim 3 is characterized in that In step S3, the cosine similarity The formula is: ； in, Indicates the i-th data point of the initial embedding of the view; represents the jth data point of the consistent embedding matrix; T represents transpose; According to the evidence pair, the belief quality and uncertainty between the initial embedding and the consistent embedding matrix are calculated, and the formula is: ； ； ； in, represents the quality of the belief; express the said uncertainty; Represents the Dirichlet distribution parameters corresponding to data point i and data point j; represents the Dirichlet intensity; N represents the number of samples; The uncertainty is modeled using the probability density function of the Dirichlet distribution, as follows: ； in, Represents a random variable The obedience parameters are The Dirichlet distribution of represents the overall reliability probability of data point i; Represents the reliability probability of the similarity between data point i and data point j, ; represents all Dirichlet distribution parameters associated with data point i; represents the N-dimensional beta function; represents the N-dimensional unit simplex; After obtaining the reliable and consistent embedding, according to the indicator matrix, the loss function and Back propagation, continuously optimizing and updating model parameters, the formula is: ； ； in, Represents the vector corresponding to the i-th data point in the indicator matrix; represents the KL divergence measure; represents the parameters of the Dirichlet distribution after removing unreliable evidence from the predicted Dirichlet distribution.

5. The multi-view clustering method based on uncertainty and graph neural network according to claim 1, characterized in that ,The S4 is specifically: taking the reliable consistent embedding and the transfer matrix as the input of the graph convolution encoder, passing through two graph convolution layers and a ReLU activation function, to obtain the final embedding.

6. The multi-view clustering method based on uncertainty and graph neural network according to claim 1, characterized in that , the S5 is specifically: The final embedding is obtained by reconstructing the initial input through the decoder to obtain the reconstructed embedding, and the mean square error loss function is used to optimize the difference between the reconstructed embedding and the reliable consistent embedding. The formula of the mean square error loss function is: ； in, represents said reliable consistent embedding; represents said reconstructed embedding; represents the Frobenius norm of the matrix; N represents the number of samples; represents the dimension of fused embedding; According to the final embedding, a clustering result is obtained by a clustering function.

7. The multi-view clustering method based on uncertainty and graph neural network according to claim 1, characterized in that , and also includes: inputting the preprocessed and labeled multi-view training set into the multi-view clustering model for training and parameter fine-tuning to obtain a trained multi-view clustering model; During the training process, the clustering loss function is used to train the clustering target, the clustering result of the multi-view clustering model is optimized, the initial probability distribution Q is calculated using the t-student distribution, and then the target probability distribution P is calculated based on the initial probability distribution Q. The formula of the clustering loss function is: ； in, represents the clustering loss function, which represents the difference between the initial probability distribution Q and the target probability distribution P; represents the KL divergence measure; and Respectively indicate that data point i belongs to cluster The target probability P and the initial probability Q.

Images